[EM] Seven +/- Two

[Bart Ingles]

I pretty much agree with all of this.  I have long thought that pairwise
methods made more sense if the number of preference levels were
restricted.  Possibly a number of levels as a function of the number of
candidates, with 2 levels (equivalent to approval voting) if there are
few (say three to five) candidates, and a maximum of three or four
levels when there are many candidates.

This has much the same effect as ranking pairs (as Mr. Ketchum points
out).

But I seriously doubt that this would have much, if any, practical
benefit over plain approval voting.  After all of the recent talk about
chads, undervoting, and overvoting, it seems ironic that the greatest
single improvement we could make to single winner elections would be to
*legalize* overvoting.

Bart



Buddha Buck wrote:
> 
> A message on another list reminded me of something -- an old,
> well-established psychology paper entitled "The Magical Number Seven,
> Plus or Minus Two: Some Limits on Our Capacity for Processing
> Information" by George A. Miller (The Psychology Review, 1956, vol 63,
> pp. 81-97, republished at http://www.well.com/user/smalin/miller.html
> by permission of author, address checked 2001-09-10).
> 
> What the article talks about is the ability of people to perceive and
> distinguish stimulae accurately.  To a very large degree, Miller noted
> that most people can easily distinguish things into about 7 plus or
> minus two groups -- we can identify 7 plus/minus two shades of grey, 7
> tones, 7 time intervals, etc.  Given multiple dimensions to
> distinguish, we can do better -- we can easily distinguish things into
> 25 partitions of a square, 30 combinations of tones on a variety of
> instruments, etc -- as long as the amount of information for each
> variable is still about 7 +/- 2.
> 
> I believe, because of this paper, and the research along those same
> lines that has followed it, that any attempt to try to get more than
> 3-4 bits of information from the voter per candidate is likely to
> result in lots of noise and voter error.  CR, while theoretically nice
> (perhaps epecially when done on a -100 to 100 scale), may run afoul of
> too much noise if the range of cardinalities greatly exceeds seven +/-
> two.  Rankings above 5 or 6 candidates becomes difficult. And so
> forth.
> 
> The ballot proposed by Forest (I believe) involving grading candidates
> A, B, C, D, and F (as in school grades) seems to me to have about the
> right information content to record the opinions of the voter with
> reasonable accuracy and precision.  Approval's ballot is simpler, and
> seems to provide enough information to make a good decision.
> 
> Just something I was thinking about.
> 
> Later,
>   Buddha